Establishing parallel trends in a probit differences-in-differences model

[Karishma DSouza]

Hi,

I have a question that's slightly more conceptual than related to direct stata output. I understand that in a probit model the coefficients in the stata output by themselves are meaningless and that it is the margins output that should be interpreted. I am running a probit differences-indifference model and wanted to know what one should look at when trying to establish the parallel trends pre-treatment. In a linear probability model, I would've plotted the average values of the residuals for the treatment and control group to establish parallel trends, but I'm not sure if that's meaningful in a probit model.

Any guidance will be much appreciated.

Thanks,
Karishma

[Joseph Coveney]

I understand that in a probit model the coefficients in the stata output by themselves are meaningless and that it is the margins output [(predicted probability estimates)] that should be interpreted.

Really? Others may disagree, but if a probit regression model is what's suitable here, then I would look to linear predictions in the estimation metric for parallel trends.

[Karishma DSouza]

Than you, Joseph.